API Endpoints

l3mcore listens on http://0.0.0.0:11435 by default and exposes endpoints compatible with the OpenAI and Ollama APIs.

---

OpenAI Endpoints

POST /v1/chat/completions

Generates a chat response. Compatible with any OpenAI client.

Request:

{
  "model": "gpt-4",
  "messages": [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Write a bash script to list processes."}
  ],
  "stream": true
}

Optional model field

• Omitted or generic (e.g. "gpt-4"): l3mcore analyzes the prompt and automatically routes to the best expert.
• Specific label (e.g. "programmer"): skips the router and goes straight to the indicated expert.

Response (without streaming):

{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "model": "programmer",
  "choices": [{
    "message": {
      "role": "assistant",
      "content": "#!/bin/bash\nps aux | sort -rk 3,3 | head -20"
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 15,
    "completion_tokens": 12,
    "total_tokens": 27
  }
}

---

GET /v1/models

Lists all available experts in OpenAI format.

Response:

{
  "object": "list",
  "data": [
    {"id": "programmer", "object": "model", "created": 1700000000, "owned_by": "lemoe"},
    {"id": "creative_writer", "object": "model", "created": 1700000000, "owned_by": "lemoe"}
  ]
}

---

Ollama Endpoints

POST /api/chat

Compatible with Ollama clients.

{
  "model": "lemoe",
  "messages": [
    {"role": "user", "content": "Explain quantum computing."}
  ],
  "stream": false
}

---

GET /api/tags

Lists experts in Ollama format.

---

GET /api/version

Returns the server version:

{"version": "1.0.0"}

---

Server Limits

Limit	Value
Rate limit	60 requests/minute per IP
Max payload size	1 MB
Streaming	SSE compatible

Behind a reverse proxy

The rate limiter respects the X-Forwarded-For header, so it works correctly behind Nginx, Traefik, or Caddy.